Lucy Farnik

Research Intern @ Jane Street | ML PhD Student | Previously: Senior Dev @ B2B SaaS; MATS; ARENA; BASC

United Kingdom

About

I'm a researcher and software engineer passionate about the frontier of machine learning. I started coding at age 7, which led to me becoming a senior developer at a startup at age 18. In my first 6 months of doing ML research, I submitted to NeurIPS as the second-name author along with researchers from Oxford, UC Berkeley and FAR AI. I've also done MATS, ARENA, AISC, and cofounded a small research organization. I'm currently a PhD student working on LLM robustness with a broad range of methods.

Experience

  • Machine Learning Research Intern at Jane Street
    Jun 2026 - Present · 2 mos

  • University of Bristol (3 yrs 1 mo)
    • PhD Researcher
      Sep 2023 - Present · 2 yrs 11 mos

      Researching how modern ML systems work under the hood in order to make them more robust. I'm supervised by Dr. Laurence Aitchison.

    • Research Intern
      Jul 2023 - Sep 2023 · 3 mos

      I am working on mechanistic interpretability of LLMs under the supervision of Dr. Conor Houghton, specifically analyzing how GPTs "understand" the difference between verbs and nouns.

  • Research Scholar at ML Alignment & Theory Scholars
    Nov 2023 - Jul 2024 · 9 mos

    Exploring circuit-style analysis with SAE features under Neel Nanda (Google DeepMind)

  • Research Lead & Cofounder at Bristol AI Safety Centre
    Aug 2023 - Jan 2024 · 6 mos

    I co-founded and am currently co-leading a small research organization focused on addressing the hard problems in AI interpretability and control, as well as investigating opportunities for more effective AI regulation. We also help upskill promising students from the University of Bristol. We received our first grant in September 2023.

  • Research Mentee at Epoch
    Jul 2023 - Sep 2023 · 3 mos

    Analyzing major paradigm shifts throughout the history of AI in order to forecast how often we can expect to see them in the coming years. My research is supervised by Pablo Villalobos in a joint program with Epoch and the Forecasting Research Institute.